Odoo 20 AI Agents: What Could They Actually Do in a South African Business?

We've been talking to customers about AI in Odoo for some time now, and one thing is becoming fairly clear.

Most businesses aren't really looking for another chatbot.

They want to know whether AI can remove some of the repetitive work their staff are doing every day.

That's why the AI agent functionality being introduced with Odoo 20 caught our attention.

The interesting part isn't that Odoo can use AI to answer a question or write some text. We've had that type of functionality for a while.

What's changing is the ability for AI to become more involved in the actual Odoo process, including working with records and information inside the system.

That has some much more practical possibilities.

What Do We Mean by an AI Agent?

Let's use a normal customer enquiry as an example.

A customer sends an email asking for information on a product. They mention what they need, quantities, delivery requirements and perhaps attach a specification.

Someone in the sales department normally has to read the email, look at the attachment, check whether the customer already exists, find the relevant products and then start capturing information into Odoo.

Using AI simply to summarise that email saves a little time.

Using AI to interpret the enquiry and help prepare the relevant information in Odoo saves considerably more.

That's where the idea of an AI agent starts becoming useful.

Odoo has demonstrated Odoo 20 AI agents being able to work with records rather than simply return text to the user.

We think that distinction matters.

Take Helpdesk as Another Example

Helpdesk is probably one of the easiest places to see where this could work.

Imagine a company receiving 500 support tickets a month.

Some tickets are two sentences long. Others contain a long email trail, screenshots, technical information and several previous conversations.

Before a consultant can solve anything, they first need to understand what has happened.

AI could potentially read that history and give the consultant something much simpler:

The customer has reported the same problem twice.

The issue appears to relate to a delivery.

The previous consultant requested additional information yesterday.

The customer has now supplied it.

There is an internal procedure covering this type of problem.

Now the consultant can get on with solving the issue.

That, to us, is a far more useful application of AI than simply asking it to write a polite email.

There is also another step.

If the agent can work with Odoo records, it could potentially help classify the ticket, update information or prepare the next action.

We'd still want the consultant involved, particularly where a decision needs to be made, but much of the administration around the process could potentially be reduced.

Manufacturing Is Where It Gets Interesting

Manufacturing businesses generate an enormous amount of information that doesn't always sit neatly in a database field.

A maintenance technician writes a note.

An operator reports a recurring problem.

Quality records a failure.

Purchasing receives an email from a supplier about a delayed component.

Production records downtime.

Individually, each piece of information may not tell you very much.

Put them together and there may be a pattern.

A production manager could ask why a particular product has repeatedly been late and AI could help work through the information available in Odoo to identify what deserves investigation.

Perhaps the same component has been short three times.

Perhaps a particular machine has had repeated maintenance problems.

Perhaps quality failures have increased.

AI shouldn't make the production decision for the manager.

But if it can save the manager an hour of searching through records to understand the problem, that's useful.

We've always felt this is where AI in ERP will eventually prove its value.

Not replacing the person who understands the business, but reducing the amount of work required to get the information that person needs.

Documents Could Be One of the Bigger Opportunities

Odoo 20's AI direction also includes closer interaction with documents.

This doesn't sound as exciting as an AI agent creating records, but in a real business it could be extremely useful.

Think about how much company knowledge sits in PDFs, procedures, product specifications, supplier documentation and technical manuals.

A user might know the answer to a question exists somewhere, but finding it is another matter.

Instead of opening documents one by one, the user could ask a question against the relevant information.

A new employee might ask:

"What is our procedure when a customer returns damaged stock?"

A technician could ask:

"What does the manufacturer recommend for this fault?"

A buyer might want to know:

"What lead time did this supplier quote in the latest documentation?"

The time saving doesn't come from AI knowing the answer.

It comes from AI finding the answer in information the business already has.

There is an important condition here: the AI should only be working with information the user is entitled to access.

That needs to be designed properly.

MCP Is Worth Watching

Odoo has also been talking about Model Context Protocol, or MCP, with Odoo 20.

It's a technical subject, but the basic idea is reasonably straightforward.

MCP provides a standard way for AI systems to connect to tools and information.

Why do we think that's important?

Because most companies aren't going to have only one AI platform.

A business may use Odoo's AI functionality for some processes, Claude for others, and perhaps ChatGPT or a specialised AI application elsewhere.

If those systems can connect to business information in more standardised and controlled ways, it opens up some interesting integration possibilities.

We're already looking at how platforms such as Claude can work alongside Odoo, so MCP is something we'll be following closely as Odoo 20 develops.

Where We'd Be Careful

The fact that an AI agent can do something doesn't mean it should be allowed to.

There is a big difference between an agent preparing a CRM activity and an agent approving a R500,000 purchase order.

We would be quite comfortable exploring AI that reads a Helpdesk ticket and prepares information for the consultant.

We'd also look at AI helping structure a sales enquiry or turning a technician's dictated notes into a proper service report.

We'd be much more cautious about allowing AI to independently post accounting transactions, change customer credit limits, adjust stock, approve payments or make significant purchasing decisions.

Odoo already has users, access rights and approval processes for a reason.

AI shouldn't become a way around them.

It should operate within them.

There is also the question of data.

If information is being sent to an external AI service, the business needs to understand what information is being sent, why it is required, how it is handled and whether personal or commercially sensitive information is involved.

For South African businesses, that means POPIA needs to form part of the discussion.

It isn't something to think about after the integration has been built.

How Would We Approach Odoo 20 AI?

We wouldn't start by switching on every AI feature we can find.

We'd choose one process.

Preferably an annoying one.

Find something employees are doing repeatedly that involves reading, searching, summarising, categorising or capturing information.

Then measure it.

If five people in a Helpdesk team are each spending an hour a day reading through ticket histories and categorising enquiries, there's something worth investigating.

If a salesperson receives three complicated enquiries a month, building an elaborate AI process around them probably isn't going to change the business.

That's the test we'd use.

There needs to be enough of a problem to justify solving it.

Start there, get the controls right and see whether the promised time saving actually appears.

If it does, move to the next process.

Our View of Odoo 20 AI

We don't think the big story with Odoo 20 is that Odoo has more AI.

Almost every business software company can say that now.

What interests us is that AI is starting to move closer to the transaction.

It's moving from writing an answer to helping with the work around that answer.

For a Helpdesk consultant, that could mean less time reading old tickets.

For sales, less time working through lengthy enquiries.

For manufacturing, it could make it easier to understand what is causing recurring problems.

For technicians, it could mean speaking their job notes rather than typing them.

None of those things replaces the person doing the job.

They remove some of the work surrounding the job.

And that's where we think South African businesses should start when looking at Odoo 20 AI.

Don't ask:

"How much AI can we put into Odoo?"

Ask:

"What are our people doing every day that they shouldn't still have to do manually?"

The answer to that question is probably where the first useful Odoo AI project will be found.

At AP Systems, we're continuing to evaluate Odoo's own AI functionality as well as how external AI platforms can work with Odoo. Our approach is to start with the business process, understand the information and controls involved, and only then decide where AI makes sense.Start writing here...